茂之钳
114 lines
3.7 KiB
C++
114 lines
3.7 KiB
C++
/// bench_spatial.cpp — R-Tree / KD-Tree build + kNN benchmarks
|
|
///
|
|
/// Metrics:
|
|
/// RTree_Build/N{size} — STR-bulk-build time for N random 3D points
|
|
/// RTree_kNN/N{size} — kNN(n=10) queries on a pre-built R-Tree
|
|
/// KDTree_Build/N{size} — build time for N random 3D points
|
|
/// KDTree_kNN/N{size} — kNN(n=10) queries on a pre-built KD-Tree
|
|
|
|
#include <benchmark/benchmark.h>
|
|
#include <vde/spatial/r_tree.h>
|
|
#include <vde/spatial/kd_tree.h>
|
|
#include <vde/core/point.h>
|
|
#include <random>
|
|
#include <vector>
|
|
|
|
using namespace vde;
|
|
|
|
namespace {
|
|
|
|
std::vector<core::Point3D> random_points_3d(int n, double R = 100.0) {
|
|
std::mt19937 rng(42);
|
|
std::uniform_real_distribution<double> dist(-R, R);
|
|
std::vector<core::Point3D> pts;
|
|
pts.reserve(n);
|
|
for (int i = 0; i < n; ++i)
|
|
pts.emplace_back(dist(rng), dist(rng), dist(rng));
|
|
return pts;
|
|
}
|
|
|
|
/// 20 random query points for kNN
|
|
std::vector<core::Point3D> query_points(int n = 20, double R = 100.0) {
|
|
return random_points_3d(n, R);
|
|
}
|
|
|
|
} // namespace
|
|
|
|
// ── R-Tree ────────────────────────────────────────────────────────────
|
|
|
|
static void RTree_Build(benchmark::State& state) {
|
|
auto pts = random_points_3d(state.range(0));
|
|
for (auto _ : state) {
|
|
spatial::RTree<core::Point3D> tree;
|
|
tree.build(pts);
|
|
benchmark::DoNotOptimize(tree.size());
|
|
}
|
|
state.SetItemsProcessed(state.iterations() * state.range(0));
|
|
}
|
|
BENCHMARK(RTree_Build)->Arg(500)->Arg(1000)->Arg(2000);
|
|
|
|
class RTreeFixture : public benchmark::Fixture {
|
|
public:
|
|
void SetUp(const benchmark::State& state) override {
|
|
points = random_points_3d(state.range(0));
|
|
tree.build(points);
|
|
queries = query_points();
|
|
}
|
|
|
|
spatial::RTree<core::Point3D> tree;
|
|
std::vector<core::Point3D> points;
|
|
std::vector<core::Point3D> queries;
|
|
};
|
|
|
|
BENCHMARK_DEFINE_F(RTreeFixture, RTree_kNN)(benchmark::State& state) {
|
|
size_t total = 0;
|
|
for (auto _ : state) {
|
|
for (const auto& q : queries) {
|
|
auto result = tree.query_knn(q, 10);
|
|
total += result.size();
|
|
}
|
|
}
|
|
benchmark::DoNotOptimize(total);
|
|
state.SetItemsProcessed(state.iterations() * queries.size());
|
|
}
|
|
BENCHMARK_REGISTER_F(RTreeFixture, RTree_kNN)->Arg(500)->Arg(1000)->Arg(2000);
|
|
|
|
// ── KD-Tree ──────────────────────────────────────────────────────────
|
|
|
|
static void KDTree_Build(benchmark::State& state) {
|
|
auto pts = random_points_3d(state.range(0));
|
|
for (auto _ : state) {
|
|
spatial::KDTree<core::Point3D> tree;
|
|
tree.build(pts);
|
|
benchmark::DoNotOptimize(tree.size());
|
|
}
|
|
state.SetItemsProcessed(state.iterations() * state.range(0));
|
|
}
|
|
BENCHMARK(KDTree_Build)->Arg(500)->Arg(1000)->Arg(2000);
|
|
|
|
class KDTreeFixture : public benchmark::Fixture {
|
|
public:
|
|
void SetUp(const benchmark::State& state) override {
|
|
points = random_points_3d(state.range(0));
|
|
tree.build(points);
|
|
queries = query_points();
|
|
}
|
|
|
|
spatial::KDTree<core::Point3D> tree;
|
|
std::vector<core::Point3D> points;
|
|
std::vector<core::Point3D> queries;
|
|
};
|
|
|
|
BENCHMARK_DEFINE_F(KDTreeFixture, KDTree_kNN)(benchmark::State& state) {
|
|
size_t total = 0;
|
|
for (auto _ : state) {
|
|
for (const auto& q : queries) {
|
|
auto result = tree.query_knn(q, 10);
|
|
total += result.size();
|
|
}
|
|
}
|
|
benchmark::DoNotOptimize(total);
|
|
state.SetItemsProcessed(state.iterations() * queries.size());
|
|
}
|
|
BENCHMARK_REGISTER_F(KDTreeFixture, KDTree_kNN)->Arg(500)->Arg(1000)->Arg(2000);
|